Introduction
Watermelon is an important crop in the United States, whose farm value is estimated at $340 million (www.watermelon.org). Economic and nutraceutical importance of this crop is rapidly increasing throughout the world. Severe bottlenecks in the genetic background of cultivated watermelon have been reported based on the DNA marker analysis of genetic similarities (Navot and Zamir, Reference Navot and Zamir1987; Zhang et al., Reference Zhang, Rhodes and Skorupska1994; Lee et al., Reference Lee, Shin, Park and Hong1996; Levi et al., Reference Levi, Thomas, Keinathand and Wehner2001, Reference Levi, Thomas, Newman, Reddy, Zhang and Xu2004).
According to Livingstone (1857), Meeuse (Reference Meeuse1962) and Pitrat et al. (Reference Pitrat, Chauvet and Foury1999), the species Citrullus lanatus (Thunb.) Matsum and Nakai originated in Kalahari region of Namibia and Botswana (Bates and Robinson, Reference Bates, Robinson, Smart and Simmonds1995; Robinson and Decker-Walters, Reference Robinson and Decker-Walters1997; Ellul et al., Reference Ellul, Lelivelt, Naval, Noguera, Sanchez, Atarés, Moreno, Corella and Dirks2007). The species C. lanatus includes two botanical varieties, namely var. lanatus (Bailey) and var. citroides (Mansf). Cultivated watermelons belong to var. lanatus and have endocarps in wide-ranging colours. The var. citroides is cultivated in southern Africa, and also called ‘Tsamma’ or ‘citron’ melon, whose rind is used as preservative in pickles (Burkill, Reference Burkill1985; Jarret et al., Reference Jarret, Merrick, Holms, Evans and Aradhya1997; Jeffrey, Reference Jeffrey and Hanelt2001). The citron fruits have green- or white-coloured flesh and their taste may vary from bland to bitter. Seed production fields should be isolated from weedy citron types since these two botanical varieties cross readily (Wehner, Reference Wehner2007). The species Citrullus colocynthis (Schrad) is a perennial herb known as bitter apple and is a desert species with a rich history as a medicinal plant (Dane et al., Reference Dane, Liu and Zhang2007). T.W. Whitaker considered C. colocynthis to be a likely ancestor of watermelon as it is morphologically similar to lanatus, is freely intercrossable and produces fertile hybrids (Wehner, Reference Wehner2007). Dane et al. (Reference Dane, Liu and Zhang2007) reported divergent lineages of colocynthis that are from tropical Asia and Africa, now widely distributed in the Saharo-Arabian phylogeographic region of Africa and in the Mediterranean region.
In earlier reports, isozyme and random amplified polymorphic DNA (RAPD) markers were used extensively in molecular diversity and phylogenetic analyses in Citrullus spp. (Zamir et al., Reference Zamir, Navot and Rudich1984; Navot and Zamir, Reference Navot and Zamir1987; Biles et al., Reference Biles, Martyn and Wilson1989; Levi et al., Reference Levi, Thomas, Keinathand and Wehner2001a, Reference Levi, Thomas and Wehnerb). Hillis (Reference Hillis and Hall1994) and Harris (Reference Harris1995) argued against the use of RAPDs in phylogenetic analysis because of their questionable homology assessments. In contrast, the simple sequence repeat (SSR) and amplified fragment length polymorphism (AFLP) markers are highly repeatable and in combination they are very efficient in resolving phylogenetic relationships, germplasm evaluation, quantitative trait locus (QTL) analysis and building integrated genetic maps (Ramakrishnan et al., Reference Ramakrishnan, Meyer, Waters, Stevens, Coleman and Fairbanks2004; Koopman et al., Reference Koopman, Wissemann and Cock2008). AFLP, in particular, has proven useful for estimating relationships among closely related species in a wide variety of lineage (Spooner et al., Reference Spooner, Mclean, Ramsay, Waugh and Bryan2005; Pimentel et al., Reference Pimentel, Sahuquillo and Catalán2007; Thaler et al., Reference Thaler, Brandstätter, Meranera, Chabicovskia, Parsonc, Zelgerd, Dallaviad and Dallinge2008).
Levi et al. (Reference Levi, Thomas, Keinathand and Wehner2001) used 662 RAPD markers to construct a dendogram with little resolution, which did not separate taxa from lanatus and citroides clearly. Later, Levi et al. (Reference Levi, Thomas, Newman, Reddy, Zhang and Xu2004) used combined analysis of AFLP and inter-SSR among heirloom watermelons and concluded that this joint analysis resolved better genetic diversity than RAPDs. Valuable phylogenetic information pertaining to the genus Citrullus was generated using chloroplast-specific polymerase chain reaction–restriction fragment length polymorphism and comparative chloroplast gene sequence data (Dane, Reference Dane2002; Dane et al., Reference Dane, Lang and Bakhtiyarova2004; Dane and Liu, Reference Dane and Liu2007). Although the resolution obtained in these studies was higher than that in RAPDs, several studies in other genera concluded that genome-wide character sets will have greater potential as phylogenetic markers (Eriksen and Töpel, Reference Eriksen and Töpel2006; Koopman et al., Reference Koopman, Wissemann and Cock2008). AFLPs were also widely used for resolving cytonuclear conflicts in several other organisms (Sullivan et al., Reference Sullivan, Lavoue, Arnegard and Hopkins2004; Kyndt et al., Reference Kyndt, Romeijnpeeters, Droogenbroeck, Romeromotochi, Gheysen and Goetghebeur2005; Schönswetter et al., Reference Schönswetter, Suda, Popp, Weissschneeweiss and Brochmann2007).
The current study aims to use AFLPs and SSRs jointly to resolve genome-wide molecular phylogenies among the same Citrullus taxa that were analyzed previously using RAPDs by Levi et al. (Reference Levi, Thomas, Keinathand and Wehner2001) and further to subject the shared polymorphisms between two taxa combinations to principal component analysis (PCA) to resolve interrelationships. We also report 30 newly developed polymorphic SSRs and relevant primer sequence information for use of the watermelon-breeding community.
Materials and methods
Plant material and DNA isolation
Seeds of 31 accessions (Table 1) were kindly provided by Dr Robert Jarret, Plant Genetic Resources Conservation Unit, USDA–ARS, Griffin, GA, 30 223. DNA was extracted from leaf tissues using the method described in the DNeasy Plant Mini Kit (Qiagen, Hilden, Germany).
Table 1 Phenotypic classifications and origin of the Citrullus spp. used in the study
* As reported in the Germplasm Resource Information Network reference base.
AFLP analysis
AFLP analysis was carried out using the protocols and kits developed by LI-COR Biosciences (Lincoln, NE, USA, www.licor.com). The EcoRI and MseI enzyme-digested products were ligated to respective restriction site (RS)-specific adapters and diluted tenfold. Diluted adapter-ligated templates were pre-amplified using adapter-specific primers with overhangs of A and C for EcoRI and MseI, respectively. Pre-amplified products were further diluted 20-fold and subjected to selective amplification using IR-700 or IR-800 labelled EcoRI-AXX primers and unlabelled MseI-CXX primers using standard touchdown PCR conditions (Vos et al., Reference Vos, Hogers, Bleeker, Reijans, van de Lee, Hornes, Fritjers, Pot, Paleman, Kuiper and Zabeau1995). A total of 35 primer combinations were used (Table 2). Amplified products were denatured and resolved on a LICOR-4500 genotyper.
Table 2 Total number of amplified and polymorphic fragments with 35 amplified fragment length polymorphism primer combinations
SSR development
Genomic DNA and fruit tissue-specific cDNA were simultaneously digested with a set of restriction enzymes. Purified, digested DNA was ligated to AP-11 (5′CTCTTGCTTAGATCTGGACTA3′) and AP-12 (5′pTAGTCCAGATCTAAGCA-AGAGCACA3′, where p = 5′ phosphate) adapters and hybridized with biotin-labelled oligos, which contain the repeat motifs. DNA fragments that are hybridized with the repeat oligos were separated using streptavidin beads. These repeat motif-enriched fragments were separated from the beads in an alkaline buffer for purification using a Qiagen PCR purification kit. These enriched fragments were cloned (TOPO cloning kit; Invitrogen, Madison, WI, USA) and 96 randomly picked clones were sequenced. The sequences with repeat motifs were identified and used for designing SSR primer pairs. PCR conditions for SSRs were used as per Reddy et al. (Reference Reddy, Pepper, Abdurakhmonov, Saha, Jenkins, Brooks, Bolek and El-Zik2001), and gel electrophoresis was carried out using super fine resolution agarose (www.amresco-inc.com).
Data scoring and analysis
Minor AFLP polymorphisms that were not uniformly amplified (e.g. were faint or not distinct in some genotypes) were eliminated from the analysis. Similarly, stutter and background bands were not considered while scoring SSR markers. The presence or absence of each fragment was scored as a binary unit character (1 = present and 0 = absent) in the case of AFLPs, and in the case of SSRs, scoring was the presence or absence of putative alleles. Genetic similarities based on Jaccard's coefficients (Jaccard, Reference Jaccard1908) were calculated using the SIMQUAL program of the Numerical Taxonomy Multivariate Analysis System (NTSYS-pc) Version 2.0 software package (Rohlf, Reference Rohlf1987). The resulting genetic similarity indices were used to generate a tree using the neighbour-joining (NJ) method (Saitou and Nei, Reference Saitou and Nei1987). PAUP*4.0 was used to generate 1000 bootstrap replicates for testing the reliability of the dataset and to draw a consensus tree. Principal component analysis (PCA) based on the genetic similarity matrices was performed using of NTSYS-pc.
Results
A range of 146–324 bands were amplified per primer combination totalling 6879 AFLP markers that were amplified from 35 primer combinations. Of these, 3089 bands (45%) were polymorphic among the 31 accessions sampled (Table 2). We counted specific bands that were polymorphic within the individual species and the shared bands between taxa. Five hundred and eighty-three amplicons were polymorphic within the taxa of lanatus, and 505 and 194 markers were polymorphic within citroides and colocynthis groups, respectively. The shared bands between any two taxa groups were also counted to determine how the shared polymorphisms will resolve interrelationships. Six hundred and fifty-two bands were shared between lanatus and citroides, 756 shared bands between lanatus and colocynthis and 620 bands between colocynthis and citroides. However, it is very difficult to identify species-specific bands using AFLP as they are dominant marker and individual bands are generally shared among species. The markers that are polymorphic within lanatus can be very important for the improvement of cultivated watermelon as they can be used for mapping and marker-assisted selection. In this study, AFLP markers were more polymorphic than SSRs within lanatus group.
Thirty SSRs amplified 169 alleles in 31 watermelon accessions. A range of 2–12 alleles were amplified per SSR. Primer sequence information and the range of amplified product sizes across the species are presented in Supplementary Table 1, available online only at http://journals.cambridge.org. The number of specific alleles was 50, 60 and 59 specific to var. lanatus, var. citroides and C. colocynthis, respectively.
To start with, diversity analysis was carried out for both AFLP and SSR separately, which interestingly resulted in similar grouping pattern of the selected accessions with the datasets when analyzed together (trees not presented). Combined AFLP and SSR analyses revealed that the genetic distances % (GDs) within the accessions of various species were 42, 38 and 34 for lanatus, citroides and colocynthis, respectively. Distance between the species groups was wider than within the species, as predicted. Our study revealed that the GD between the lanatus and citroides taxa was 40, and the GD between the species colocynthis and citroides was 43. The GD between lanatus and colocynthis was estimated to be 46, indicating that the var. citroides is closely related to the species colocynthis than var. lanatus.
A NJ phenogram was constructed using the combined datasets of AFLPs and SSRs (Fig. 1). Trees were rooted using the taxa C. colocynthis as the out-group. A monophyletic cluster of colocynthis was basally resolved with a bootstrap value support of 95. The second branch of the tree was of the species C. lanatus, which further split into two sister clades of var. citroides and var. lanatus. This split was supported with a breeding value (BV) of 97. The lanatus cluster had six lanatus subclusters, which were closely resolved in the middle of the tree with a BV range from 57 to 82. The sixth subcluster of lanatus was an exception in the whole analysis as two of the citroides' taxa from southern Africa were grouped into this subcluster. A well-supported monophyletic cluster composed of seven citroides accessions is resolved on top of the tree.
Fig. 1 NJ phenogram of 31 Citrullus spp. using AFLPs and newly captured SSRs. Numbers shown at different nodes represent percentage confidence limits obtained in the bootstrap analysis.
We separated the data into four subsets, which were polymorphic within different groups of (1) lanatus vs. citroides (Fig. 2), (2) lanatus vs. colocynthis (Fig. 3) and (3) citroides vs. colocynthis (Fig. 4). When PCA was done using the dataset of lanatus alone (figure not presented), the first three eigen vectors absorbed 36.90, 17.78 and 12.48 totalling 67.16% of total variation, indicating the robustness of the dataset and reliability of the analysis. The lanatus accessions PI 271778 (S. Africa), PI 169289 (Turkey), PI 248178 (central Africa), PI 169290 (Turkey) and PI 165451 (Mexico) were clustered together with the citroides' taxa (cluster II), and the rest of the 11 lanatus taxa grouped as cluster I (Fig. 2). These lanatus' and citroides' PIs from cluster I may have been transitional between these two taxa. When we combined the dataset of lanatus with citroides, the PCA revealed that cluster I from the lanatus-specific PCA clustered with the six citroides types (Fig. 2). This PCA was supported by eigen vectors I (31.23), II (16.40) and III (9.75), cumulatively explaining 57.38% of the total possible variation. The PCA with lanatus and colocynthis explained about 65.63% of the total variation (eigen vectors I = 33.57, II = 19.58 and III = 12.48), and also reveals the clustering of colocynthis types with the cluster II from the lanatus-specific PCA (Fig. 3).
Fig. 2 Three-dimensional picture of principal component analysis estimated using AFLP and SSRs' genetic similarity matrix of 16 C. lanatus var. lanatus and 9 C. lanatus var. citroides' accessions.
Fig. 3 Three-dimensional picture of principal component analysis estimated using AFLP and SSRs' genetic similarity matrix of 16 C. lanatus var. lanatus and 6 C. colocynthis' accessions.
Fig. 4 Three-dimensional picture of principal component analysis estimated using AFLP and SSRs' genetic similarity matrix of nine C. lanatus var. citroides and six C. colocynthis' accessions.
To understand the species relationship between the taxa colocynthis and citroides, we further separated the dataset of these two species and subjected them to PCA separately (Fig. 4). The analysis revealed that the first three eigen vectors that were used to construct the multidimensional spectrum contributed 32.37, 25.47 and 14.57% of variation with a total of 72.42% of the total possible variance. The pattern of clustering from this PCA was congruent with the results of GD analysis, confirming that these two species are closely interrelated.
Discussion
In the current study, accessions belonging to taxa citroides, lanatus and colocynthis were clearly resolved into three well-supported clusters that are consistent with the classical taxonomic nomenclature. The accessions in the current study are from various wide geographic areas drawn from the countries Afghanistan, Argentina, China, Cyprus, Egypt, Ghana, Iran, Morocco, Namibia, Nigeria, Philippines, South Africa and Zaire. Irrespective of their geographical origin, in NJ analysis, all accessions were grouped with one of the three species-specific clusters. Geographic differentiation of genetic diversity of the genus Citrullus might have been considerably weakened by increased human interchange of cultivated watermelon and its associated weed species (citroides and colocynthis) among the countries in the past several years.
Some accessions of lanatus in the USDA-ARS germplasm show particular phenotype usually known as ‘egusi’ seed types. The egusi watermelon is commonly cultivated in Ghana, Nigeria and Congo, where the protein- and carbohydrate-rich seeds are used as a regular part of the human diet and fruits as cattle feed. Sometimes these fruit types are confused with C. colocynthis type, but they are cultivated watermelons (Wehner, Reference Wehner2007). In the current study, both of the egusi types were clustered with the lanatus group.
The current study indicated that the joint analysis of AFLP and SSR is very effective for phylogenetic analysis of Citrullus spp. Studies such as Špunarová et al. (Reference Špunarová, Ovesná, TvarŮžek, Kučera, Špunar and Hollerová2005) in barley, Gillaspie et al. (Reference Gillaspie, Hopkins and Dean2005) in Vigna, Maluf et al. (Reference Maluf, Silvestrini, Ruggiero, Filho and Colombo2005) in Coffea, Saini et al. Reference Saini, Jain, Jain and Jain2004 in rice, Perumal et al. (Reference Perumal, Krishnaramanujam, Menz, Katilé, Dahlberg, Magill and Rooney2007) in sorghum are few examples of copious published works, where joint analysis of AFLP and SSR datasets proved that these markers analyzed together have unprecedented utility for phylogenetic studies. Instead of generating a particular sequence-specific tree that does not necessarily reflect the true species tree, especially among the closely related and potentially interbreeding species such as Citrullus spp., where reticulate evolution might have occurred, the simultaneous analysis of many loci representing the whole genome has the potential to generate true phylogenies. This is an advantage of the genotyping techniques that scan diversity at loci across the genome, as the accuracy of measurements on GDs increases with the number of loci used (Travis et al., Reference Travis, Maschinski and Keim1996; Schmidt and Jensen, Reference Schmidt and Jensen2000). Our study indicated that the GD between lanatus and colocynthis is wider than the GD between the species citroides and colocynthis. These results are consistent with the GDs estimated by Levi et al. (Reference Levi, Thomas, Keinathand and Wehner2001).
SSRs are simple to use multiallelic and co-dominant marker systems that are sequence based and produce highly repeatable amplifications. The SSRs in this study generated important diagnostic markers that are species specific and can be of immense use for resolving species conflicts that are reported to exist between lanatus and citroides. Dane et al. (Reference Dane, Lang and Bakhtiyarova2004) identified diagnostic markers using cpDNA haplotypes for lanatus and citroides types and used them to track lineages with Citrullus rehmii and Citrullus ecirrhosus. Jarret et al. (Reference Jarret, Merrick, Holms, Evans and Aradhya1997) developed seven SSRs and used them to amplify 32 watermelon genotypes; they found that SSR-derived polymorphisms are very efficient in discriminating among various species. Guerra-Sanz (Reference Guerra-Sanz2002) identified 19 SSRs from cDNA sequence data. The AFLP technique is highly efficient in detecting polymorphisms at RSs and flanking sequences around RS by digestion and PCR amplification. The advantage of this technique is that AFLP can simultaneously amplify several polymorphic loci without requiring prior sequence information. In the current study, 18.8% of polymorphic markers that were generated using AFLP were informative within the cultivated watermelons. These markers could be of immense use in genetic mapping, QTL location and marker-assisted programs. Despite their advantages, SSRs and AFLPs can generate only binary character states (0–1), and hence they can introduce homoplasy into datasets. AFLP and SSR datasets are analyzed phenetically rather than cladistically. The lack of resolution is less likely to result from homoplasies than from the fact that the most of these polymorphisms are not species specific, i.e. there is retention of ancestral polymorphisms in derived lineages. However, in the current study, the tree topologies are well supported and in congruence with the tree topologies generated using cladistic methods by Dane and Liu (Reference Dane and Liu2007) and Dane and Lang (Reference Dane and Lang2004).
Evolutionary forces such as population bottlenecks, genetic drift through founder effects, adaptive radiation and recurrent gene flow due to cross compatibility between the species might have contributed to the genetic variation (Ellstrand et al., Reference Ellstrand, Prentice and Hancock1999; Dane and Lang, Reference Dane and Lang2004). Various species in the genera of Citrullus spp. are freely crossable (Wehner, Reference Wehner2007). The shared polymorphisms that were subjected to PCA further resolved interrelationships at the molecular level and the distribution of genetic lineages among these freely interbreeding taxa. These data can provide insights into different factors that shape genetic diversity (Avise, Reference Avise2000). In this paper, we resolved population structure in the taxa of lanatus. For NJ analysis, we used entire data that were generated along with a broad category of taxa including colocynthis. Clearly, when applied such a wide-ranging dataset that is not entirely informative within the lanatus cluster, the essential phylogenetic signals that are needed to resolve population substructures within lanatus cluster are shrouded. This is because only 18.8% of the total data is informative within the lanatus group and when used along with the total data for analysis, population structure within the lanatus group is masked due to sampling error in the selection of markers. To overcome this, we did separate PCAs, first with the dataset that showed polymorphism within the lanatus group, and then two additional PCAs using the shared polymorphisms with citroides and colocynthis separately to understand genetic reticulation and introgression histories. In all three PCAs (PCA-1, PCA-2 and PCA-3; lanatus itself, lanatus vs. citroides and lanatus vs. colocynthis), the taxa of lanatus grouped into two clusters. Clustering patterns in PCA-2 and PCA-3 analyses indicated that a subcluster of lanatus accessions had undergone introgression and gene flow with the both citroides and colocynthis taxa. The first cluster in both the PCAs had 11 lanatus types from northern Africa (Ghana, Nigeria), USA, Argentina and China, whereas the other five accessions of lanatus were from southern Africa, Turkey. The accessions from the Philippines and Mexico were grouped as the second cluster. Since Africa is the centre of origin for the cultivated watermelon, our results indicate that there are two predominant groups of lanatus. One group is from northern Africa (Nigeria and Ghana) and it spread across the world; the second group from the central (Zaire) and southern Africa, which introgressed considerably with the taxa citroides and colocynthis. However, a separate study should be undertaken with the large number of lanatus collections. PCA of citroides and colocynthis suggests that citroides' types have undergone extensive introgression with colocynthis, and citroides species are genetically more closely related to colocynthis than the species lanatus. The presence of divergent lineages in C. colocynthis, which form different clusters in the PCA involving taxa of colocynthis, is congruent with the results of cpDNA studies by Dane et al. (Reference Dane, Liu and Zhang2007).
Knowledge regarding the path of domestication, however, is fragmentary and various scenarios have been proposed for the origin of the domesticated watermelon from its progenitor, wild C. lanatus (Bates and Robinson, Reference Bates, Robinson, Smart and Simmonds1995; Robinson and Decker-Walters, Reference Robinson and Decker-Walters1997; Maggs-Kölling et al., Reference Maggs-Kölling, Madsen and Christiansen2000). Landraces of the Kalahari region of southern Africa (Taylor, Reference Taylor, Wickens, Goodin and Field1985) are early forms of domestication, and several others (Mallick and Masui, Reference Mallick and Masui1986) have proposed that the domestication process might also have occurred in northern Africa. Our results support both views, i.e. the lanatus accessions in the current study are two distinct groups, one being from the northern and other from central and southern Africa.
Our phylogenetic study could have been completed if we had included some accessions of C. rehmii and C. ecirrhosus, as these have been shown to be direct descendents of the species C. colocynthis (Dane and Lang, Reference Dane and Lang2004; Dane and Liu, Reference Dane and Liu2007). We made crosses between lanatus accessions from different clusters and developing recombinant inbred line populations, so that we can map the lanatus-specific polymorphisms that we identified in the current study. This might broaden the narrow genetic diversity, which is currently a bottleneck for watermelon improvement (Levi et al., Reference Levi, Thomas, Keinathand and Wehner2001, Reference Levi, Thomas, Newman, Reddy, Zhang and Xu2004). Our future endeavor is to study a large number of C. lanatus var. lanatus for resolving population structure using model-based (Bayesian) clustering algorithms (Pritchard et al., Reference Pritchard, Stephens and Donnelly2000; Falush et al., Reference Falush, Stephens and Pritchard2003; Falush et al., Reference Falush, Stephens and Pritchard2007), as Bayesian inference has proven to be useful in resolving population structure and species boundaries in several plant species that are closely related and intermating (Koopman et al., Reference Koopman, Wissemann and Cock2008).
Acknowledgements
The authors are grateful to Dr Jarret, Plant Genetic Resources Conservation Unit, USDA-ARS, Griffin, GA, 30 223, for providing the seeds of germplasm accessions. The authors thank Drs Wolfe and Collins for their critical comments. Funding support is provided by USDA-CSREES Research (2007-38 814-18 472 and 2004-38 814-15 118).
